Behind the scenes: How we trained VitoCV's AI to beat ATS filters 🤖
Since a few of you asked about the AI generation quality in the comments, I wanted to share a quick behind-the-scenes look at how the models behind VitoCV actually work. 🧠⚙️
When I started building this, I realized standard LLMs are often too generic or 'fluffy' for resumes. To fix this, I focused heavily on prompt engineering and fine-tuning the AI specifically against real-world ATS constraints and FAANG resume guidelines.
Specifically, the AI is trained to:
Prioritize Action Verbs: It automatically restructures sentences to start with strong, impactful action verbs.
Quantify Results: It actively looks for metrics in your input (or prompts you for them) to format points in the classic 'Accomplished [X] as measured by [Y], by doing [Z]' format.
ATS Keyword Mapping: It cross-references your bullet points with standard industry keywords so the ATS parsers don't filter you out.
It took a lot of iterations and testing against common ATS parsers to get the balance right—making it sound professional but still genuinely human. I’d love for you to test out the bullet point generation and let me know how it compares to your manual rewrites!
Replies